formation of dwarf galaxies in CDM cosmology (was: Kepler's Recent Results in Contradiction to Standard Planet Formation Theory)

From: "Jonathan Thornburg [remove -animal to reply]"
Subject: formation of dwarf galaxies in CDM cosmology (was: Kepler's Recent Results in Contradiction to Standard Planet Formation Theory)
Newsgroups: sci.astro.research
References:









[[I've changed the subject line, as this discussion no longer has much
of anything to do with Kepler's results or planetary formation theory]]

Robert L. Oldershaw wrote:
In the Jan. 14th issue of Nature is a paper that claims to resolve a
serious problem that has plagued the CDM cosmology for a long time.
[[...]]

For the benefit of others, the paper in question is this one:

F. Governato, C. Brook, L. Mayer, A. Brooks, G. Rhee, J. Wadsley,
P. Jonsson, B. Willman, G. Stinson, T. Quinn & P. Madau
"Bulgeless dwarf galaxies and dark matter cores
from supernova-driven outflows"
Abstract:
For almost two decades the properties of `dwarf' galaxies have
challenged the cold dark matter (CDM) model of galaxy formation1.
Most observed dwarf galaxies consist of a rotating stellar disk2
embedded in a massive dark-matter halo with a near-constant-density
core3. Models based on the dominance of CDM, however, invariably
form galaxies with dense spheroidal stellar bulges and steep central
dark-matter profiles4, 5, 6, because low-angular-momentum baryons
and dark matter sink to the centres of galaxies through accretion
and repeated mergers7. Processes that decrease the central density
of CDM halos8 have been identified, but have not yet reconciled
theory with observations of present-day dwarfs. This failure is
potentially catastrophic for the CDM model, possibly requiring a
different dark-matter particle candidate9. Here we report hydrodynamical
simulations (in a framework10 assuming the presence of CDM and a
cosmological constant) in which the inhomogeneous interstellar
medium is resolved. Strong outflows from supernovae remove
low-angular-momentum gas, which inhibits the formation of bulges
and decreases the dark-matter density to less than half of what it
would otherwise be within the central kiloparsec. The analogues of
dwarf galaxies-bulgeless and with shallow central dark-matter
profiles-arise naturally in these simulations.
Nature volume 463, number 7278, pages 203-206 (14 Jan 2010)
http://www.nature.com/nature/journal...ture08640.html
doi:10.1038/nature08640
preprint (open-access!) at http://arxiv.org/abs/0911.2237

Nature also has an "Editor's summary" and a "News and Views" article
about this work,
Marla Geha
"Galaxy formation: Gone with the wind?"
Nature volume 463, number 7278, pages 167-168 (14 Jan 2010)
http://www.nature.com/nature/journal...l/463167a.html
doi:10.1038/463167a


So now international teams of theorists using "millions of hours on
supercomputers" have run SIMULATIONS that reproduce the desired
phenomenon. Just so! Break out the champagne! Mission Accomplished!

But does anyone else see some reasons for very serious misgivings
here?

(1) The "correct" answer was assumed to be known from the start.

(2) Theorists were not going to give up until they got the "right"
answer.

I think there may be some misunderstanding here of the role that
computers and computer simulations play in theoretical astrophysics.
The usual way things get done is this:
1. Someone suggests a physical model (which usually has a bunch of
parameters). This might be based purely on speculation, and/or it
might be based on analysis of past observatious. In fact, it might
be the same as some other model which is already well-known, but
this time it's elaborated a bit more.
2. Are people able to solve the model's equations analytically?
2a: Yes -- Goto step 3 to explore the analytical solution
2b: No -- (i) Computational people figure out how to solve
the equations numerically (or, if this has already
been done, how to solve them more efficiently,
accurately, and/or robustly than has been done in
the past). If this is a lot of work, write a
paper on just this and submit it to arxiv and a journal.
(ii) Goto step 3 to explore the numerical solution.
3. Try varying the model parameters to see if we can get reasonable
agreement between the model equations' solution (either analytical
or numerical) and past observations (if there are any).
4. Try varying the model parameters some more to see if any interesting
predictions can be made about what future observations might show.
5. Write papers on steps 3 and 4 & submit them to arxiv and journals.
(Papers on step 4 also serve as hints to our observational colleagues
that they might try making said observations.)

Notice that very little of this process (only steps 2a vs 2b) depends
on whether computers are used. As Forman Acton has famously said (in
his classic (though now rather dated) book "Numerical Methods that Work"),
"the purpose of computing is insight, not numbers".

Notice also that the process is actually very similar regardless of
whether the observations predate or postdate the theoretical calculation.


(3) Using computers and many adustable parameters you can get whatever
you want.

I'm sorry, but that's simply not true. It may be that for *some*
physical systems (the models for which might have analytical solutions,
or they might not) with many adjustable parameters, you can get a wide
range of qualitative behavior. But that's certainly not the case for
all physical systems, or even for all physical systems whose equations
need a computer to solve.

A case in point: Consider a bound binary black hole system in an
otherwise empty asymptotically flat spacetime, with dynamics given
by general relativity. This system has around 15 parameters, and
simulating its outcome takes a lot of computer time (on the order
of 10,000 - 100,000 CPU-hours for a medium-resolution simulation).
But despite that, (I hereby assert) *no* combination of those
parameters will result in the binary becoming unbound. Indeed,
*no* combination of those parameters will result in any other final
end state than (the two black holes merging to form) a single black
hole.


(2) Theorists were not going to give up until they got the "right"
answer.

What the paper in question actually says is that
(a) Here's a well-known problem (a bunch of observations which weren't
well explained by past theoretical predictions).
(b) Here's a particular theoretical model (which the authors argue is
better than past models, in that it more accurately approximates
what we believe to be the underlying physics than past models did).
(c) Here are some (numerical) calculations of that model's outcomes
for certain (given) sets of parameters which (the authors assert)
are plausible.
(d) The calculated outcomes agree pretty well with past observations.

Could you be more specific about which of (a), (b), (c), and/or (d)
in this paper you consider to be not "right"?

Or, if you don't find fault with any of (a), (b), (c), or (d) in this
paper, could you explain more clearly just why it is that we should
disbelieve the author's conclusions?

Do you have any specific evidence that "you can get whatever you want"
by adjusting the paremeters in the authors' model? Is there some other
reason why we should distrust their model?

ciao,

--
-- "Jonathan Thornburg [remove -animal to reply]"
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
"Washing one's hands of the conflict between the powerful and the
powerless means to side with the powerful, not to be neutral."
-- quote by Freire / poster by Oxfam

On Jan 15, 3:30*pm, Jonathan Thornburg
wrote:

Do you have any specific evidence that "you can get whatever you want"
by adjusting the paremeters in the authors' model? *Is there some other
reason why we should distrust their model?


Forgive this very brief and partial [] response, but I need to read
your post slowly tonight with my happy endings cigar and respond more
thoughtfully and fully tomorrow.

For now I would just say that we should always maintain a healthy
skepticism when dealing with any model-building theory, for the
obvious reasons that I have already discussed in the last few days. I
repeat my von Neumann paraphrase: "One thinks of Johnny von Neumann's
dictum that with 3 variables he could produce an elephant and with 4
variables he could get it to wag its tail. Or something to that
effect." He was being humorous, but he was also saying something quite
serious.

In haste,
RLO
www.amherst.edu/~rloldershaw

On Jan 16, 3:52*am, "Robert L. Oldershaw"
wrote:
On Jan 15, 3:30*pm, Jonathan Thornburg
wrote:


Ok, I have had a chance to look over your comments, and I agree this
is paper not a "Higgs In Space" pseudoscience paper, but still it
raises the definite concerns noted below.

(A) Your #1 to #5 list of steps defines the basic strategy of the
"Model-Building" approach to physics. I do not believe that this is
the best approach to doing physics/science [see Einstein's oeuvre for
an elegant and thorough demonstration of how science can be done
better].

Sometimes "Model-Building" is the only practical route to go initially
and so it can be very useful. HOWEVER, it should be carefully guided
and tested by Definitive Predictions at all steps/stages, to the full
extent possible. Those who get intoxicated with a model and come to
regard it as received wisdom which no longer needs to be challenged
empirically, and must be "saved" whatever it takes, are fooling
themselves and potentially leading others on a false path. Need I
mention the "string theory" fiasco?

(B) " (3) Using computers and many adustable parameters you can get
whatever
you want.

I'm sorry, but that's simply not true."

I beg to differ, but I will compromise a bit and say one can get
"nearly anything" one wants badly enough.


(C) This paper is a classic "Save The Phenomenon" paper. That is not
to say that it is necessarily a bad paper, or that its results are
necessarily wrong. But let's be honest about the motivations of this
paper. Give it some objective thought and view the situation from the
full historical perspective. You will see that this is a classic STP
paper.


(D) "Or, if you don't find fault with any of (a), (b), (c), or (d) in
this
paper, could you explain more clearly just why it is that we should
disbelieve the author's conclusions?"

It could not be put more simply: they are clearly trying to validate a
predetermined desiderata. If this does not cause you to have some
serious misgivings, then I am worried for science.


(E) "Do you have any specific evidence that "you can get whatever you
want"
by adjusting the paremeters in the authors' model? "

von Neumann knew better than either of us; I let him make the case for
me.


Best,
RLO
www.amherst.edu/~rloldershaw

Robert L. Oldershaw schrieb:
Forgive this very brief and partial [] response, but I need to read
your post slowly tonight with my happy endings cigar and respond more
thoughtfully and fully tomorrow.

As this promises to become the "fight" of the fractal
cosmology paradigm against the "rest of the world", i urge
all proponents to look for a common scientifuc basis and
then advance to the controversial issues. Otherwise the
thread wil be a long medieval "Hauen und Stechen", where
each side claims the victory. This might be pleasing and
full of honour for the fighters but boring for the readers.
Regards Jurgen

[Mod. note: Quoted text deleted. In particular, if the thread turns
into an argument about one participant's personal ideas, it may be
more appropriate to take it to private e-mail -- mjh]

In article , "Robert L.
Oldershaw" writes:

(A) Your #1 to #5 list of steps defines the basic strategy of the
"Model-Building" approach to physics. I do not believe that this is
the best approach to doing physics/science [see Einstein's oeuvre for
an elegant and thorough demonstration of how science can be done
better].

As a reply to this, let me quote you in another thread:

because you say so? Is that your version of scientific falsifiation/
verification tests: whether or not something agrees with your
assumptions/biases? Really, sir!

I'm sorry, but that's simply not true."

I beg to differ, but I will compromise a bit and say one can get
"nearly anything" one wants badly enough.

because you say so? Is that your version of scientific falsifiation/
verification tests: whether or not something agrees with your
assumptions/biases? Really, sir!

For some adjusted-after-the-fact definition of "nearly". What was that
about being able to adjust things until one gets what one wants?

It could not be put more simply: they are clearly trying to validate a
predetermined desiderata. If this does not cause you to have some
serious misgivings, then I am worried for science.

You obviously misunderstand science. The "desiderata" are observations.
Obviously, if a model doesn't fit observations, it is wrong, or at least
incomplete. This is what one expects to find as observations become
better and better. So, the model is modified in that it is made more
realistic, based on physical assumptions independent of the
observations. Then, one checks if the refined model gives a better fit.
What's the problem?

Robert L. Oldershaw wrote:
Sometimes "Model-Building" is the only practical route to go initially
and so it can be very useful.

Quite true. And sometimes it is the *only* practical route to go --
some systems are simply too complicated for simple "back of the envelope"
modelling to get very far.

For example, consider operational weather forecasting: There is no
known way to produce a 5-to-8-day weather forecast that's significantly
more accurate than climatology (i.e. than simply "forecasting" the
long-term average conditions for the current location and time-of-year)
for most populated parts of the Earth without having a detailed model
of the dynamical and thermodynamical state & time-evolution of the
Earth's atmosphere. And such a such a model is *way* too complicated
to make-predictions-from without computers.

[How did weather forecasts get done before computers?
The short answer is that forecasts were a lot less accurate
back then, indeed, and that 5-to-8-day forecasts with
better-than-climatology accuracy didn't exist. See
Joseph J. Tribbia and Richard A. Anthes
"Scientific Basis of Modern Weather Prediction"
Science v.237 (31 July 1987), 493-499.
doi: 10.1126/science.237.4814.493
http://www.sciencemag.org/cgi/conten...n/237/4814/493
for more details.]


HOWEVER, it should be carefully guided
and tested by Definitive Predictions at all steps/stages, to the full
extent possible.
[[...]]
Those who get intoxicated with a model and come to
regard it as received wisdom which no longer needs to be challenged
empirically, and must be "saved" whatever it takes, are fooling
themselves and potentially leading others on a false path.

I emphatically agree with both of these statements.

--
-- "Jonathan Thornburg [remove -animal to reply]"
Dept of Astronomy, Indiana University, Bloomington, Indiana, USA
"Most investment bankers' [...] idea of a long-term investment
is thirty-six hours" -- Robert Townsend, "Up the Organization"

In article , "Robert L.
Oldershaw" writes:

As I have also mentioned before, theories of principle are the
preferred scientific strategy because they can be definitively tested,
whereas in model-building one can "cook" the statistics, the
reasoning, the assumptions, the data "cuts", the adjustable
parameters, etc., etc.,... and therefore empirical testing of model-
building results is far less certain. Yes, we get a successively
better fit to observations, but do we get a Ptolemaic fit or a
Galilean fit? To me there is a huge difference, although some
Platonists may not care one way or the other.

A counter-example from recent history: there was a HUGE prejudice in
favour of the Einstein-de Sitter model of the universe, but now the
consensus is that it is ruled out by many lines of evidence. Why isn't
there even one serious science who is a good enough cook to still use
this model as his hypothesis?